Trustworthy memory should be able to explain itself.
OpenClawBrain is built around proof surfaces, not vague personalization claims. You can inspect what memory was used, what was ignored, why something was withheld, and what the system learned from the outcome.
Tiny proof loop
openclaw plugins inspect openclawbrain --runtime
openclaw doctor
# /plugins/openclawbrain/proof?limit=10
# /plugins/openclawbrain/search?query=pnpm&limit=10
# /plugins/openclawbrain/route-policy
# /plugins/openclawbrain/graph/health
# /plugins/openclawbrain/graph/dry-runWhat you can inspect
Status
Plugin load, mode, routes, and recent runtime behavior.
Doctor
SQLite and FTS health under the current Node runtime.
Proof events
Recent route, capture, rejection, and injection decisions.
Search and graph
Local memory search, redacted nodes, and edges.
Graph maintenance
Health metrics, dry-run proposals, lineage, edge observations, and redacted proof for applied graph mutations.
Route policy
Active v3 snapshot, fallback lineage, shadow decisions, eval cases, calibration examples, and candidate reports.
The route-policy surface is the public proof of the learning loop: it shows which policy is active, whether v3 or fallback routing served, and what evidence exists before any candidate becomes production routing.
The README has the full authenticated route notes and debugging path.